On Thu Jan 22 2004 at 23:02:07 +0900, Christopher SEKIYA wrote:
> > The X driver isn't very thorough in setting up things on the newport, it
> > assumes some things are already set up "right" for it.
> 
> Understood.  Need to make both kernel driver and X driver set things up
> completely.  We're close.

I made the kernel driver do the setup as explicitly as possible.

Currently the only problem is the VC2 video mode setup, as the docs
don't delve into how the mode timing tables should be generated (on SGI
machines it's assumed those are (pre-)generated and you just load the
timings you want).
Though I had the idea of dumping the mode table from 1024x768@60Hz and
using that as fallback if I detect the card hasn't been set up at all by
ARCS, so it would always work to some extent at all...

> newport0 at gio0 slot 2 addr 0x1f000000: SGI NG1 (board revision 6)
> 
> Much, much better :)

Alright, I'll commit it.

> > Yah, but I *think* the X driver could lose some of the gfifo drains if
> > we were using uncached access - not sure though.
> 
> What I'm running seems to be very inefficient ... let me instrument things
> up and see if we gain any benefit.

It cannot be all that efficient anyway, because we don't have CPU
addressable framebuffer. What the XFree86 driver does (to keep things
simple'n'easy) is to use shadowfb and CPU-blit that to newport (since
there are no VDMA docs around :/) which, as you can guess, is anything
but fast. If we could get VDMA running it could be much faster. It could
also be even faster if we implemented everything natively in the newport
driver instead of using shadowfb, but as you can guess that's pretty big
job...

-- 
Ilpo Ruotsalainen - <lonewolf@iki.fi> - http://www.iki.fi/lonewolf/